Many facts about aphasic and nonaphasic naming are explained by models that use spreading activation to map from the semantics of a word to its phonology. The implemented model of picture naming discussed here achieves this by coupling interactive feedback with two selection steps. The model's structure and default parameters were set up to match the normal naming profile; and aphasic naming is simulated by altering ''lesionable'' parameters away from default settings. Past studies within this framework have used different sets of lesionable parameters. Here, in the largest and most representative case-series ever modeled, we show the superiority of the version of the model that allows lesions to weaken semantic and/or phonological connections. Pairing this approach to lesions with the assumptions of interactivity and two-step selection equips the model to explain the wide variation in individual naming response profiles and key facts about lexical and sublexical errors.